projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: trace old block address for CoWed page
[deliverable/linux.git] / fs / f2fs / data.c
1 /*
2 * fs/f2fs/data.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/buffer_head.h>
14 #include <linux/mpage.h>
15 #include <linux/writeback.h>
16 #include <linux/backing-dev.h>
17 #include <linux/pagevec.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
21 #include <linux/uio.h>
22 #include <linux/cleancache.h>
23
24 #include "f2fs.h"
25 #include "node.h"
26 #include "segment.h"
27 #include "trace.h"
28 #include <trace/events/f2fs.h>
29
30 static void f2fs_read_end_io(struct bio *bio)
31 {
32 struct bio_vec *bvec;
33 int i;
34
35 if (f2fs_bio_encrypted(bio)) {
36 if (bio->bi_error) {
37 f2fs_release_crypto_ctx(bio->bi_private);
38 } else {
39 f2fs_end_io_crypto_work(bio->bi_private, bio);
40 return;
41 }
42 }
43
44 bio_for_each_segment_all(bvec, bio, i) {
45 struct page *page = bvec->bv_page;
46
47 if (!bio->bi_error) {
48 SetPageUptodate(page);
49 } else {
50 ClearPageUptodate(page);
51 SetPageError(page);
52 }
53 unlock_page(page);
54 }
55 bio_put(bio);
56 }
57
58 static void f2fs_write_end_io(struct bio *bio)
59 {
60 struct f2fs_sb_info *sbi = bio->bi_private;
61 struct bio_vec *bvec;
62 int i;
63
64 bio_for_each_segment_all(bvec, bio, i) {
65 struct page *page = bvec->bv_page;
66
67 f2fs_restore_and_release_control_page(&page);
68
69 if (unlikely(bio->bi_error)) {
70 set_bit(AS_EIO, &page->mapping->flags);
71 f2fs_stop_checkpoint(sbi);
72 }
73 end_page_writeback(page);
74 dec_page_count(sbi, F2FS_WRITEBACK);
75 }
76
77 if (!get_pages(sbi, F2FS_WRITEBACK) && wq_has_sleeper(&sbi->cp_wait))
78 wake_up(&sbi->cp_wait);
79
80 bio_put(bio);
81 }
82
83 /*
84 * Low-level block read/write IO operations.
85 */
86 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
87 int npages, bool is_read)
88 {
89 struct bio *bio;
90
91 bio = f2fs_bio_alloc(npages);
92
93 bio->bi_bdev = sbi->sb->s_bdev;
94 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
95 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
96 bio->bi_private = is_read ? NULL : sbi;
97
98 return bio;
99 }
100
101 static void __submit_merged_bio(struct f2fs_bio_info *io)
102 {
103 struct f2fs_io_info *fio = &io->fio;
104
105 if (!io->bio)
106 return;
107
108 if (is_read_io(fio->rw))
109 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
110 else
111 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
112
113 submit_bio(fio->rw, io->bio);
114 io->bio = NULL;
115 }
116
117 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
118 struct page *page, nid_t ino)
119 {
120 struct bio_vec *bvec;
121 struct page *target;
122 int i;
123
124 if (!io->bio)
125 return false;
126
127 if (!inode && !page && !ino)
128 return true;
129
130 bio_for_each_segment_all(bvec, io->bio, i) {
131
132 if (bvec->bv_page->mapping) {
133 target = bvec->bv_page;
134 } else {
135 struct f2fs_crypto_ctx *ctx;
136
137 /* encrypted page */
138 ctx = (struct f2fs_crypto_ctx *)page_private(
139 bvec->bv_page);
140 target = ctx->w.control_page;
141 }
142
143 if (inode && inode == target->mapping->host)
144 return true;
145 if (page && page == target)
146 return true;
147 if (ino && ino == ino_of_node(target))
148 return true;
149 }
150
151 return false;
152 }
153
154 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
155 struct page *page, nid_t ino,
156 enum page_type type)
157 {
158 enum page_type btype = PAGE_TYPE_OF_BIO(type);
159 struct f2fs_bio_info *io = &sbi->write_io[btype];
160 bool ret;
161
162 down_read(&io->io_rwsem);
163 ret = __has_merged_page(io, inode, page, ino);
164 up_read(&io->io_rwsem);
165 return ret;
166 }
167
168 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
169 struct inode *inode, struct page *page,
170 nid_t ino, enum page_type type, int rw)
171 {
172 enum page_type btype = PAGE_TYPE_OF_BIO(type);
173 struct f2fs_bio_info *io;
174
175 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
176
177 down_write(&io->io_rwsem);
178
179 if (!__has_merged_page(io, inode, page, ino))
180 goto out;
181
182 /* change META to META_FLUSH in the checkpoint procedure */
183 if (type >= META_FLUSH) {
184 io->fio.type = META_FLUSH;
185 if (test_opt(sbi, NOBARRIER))
186 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
187 else
188 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
189 }
190 __submit_merged_bio(io);
191 out:
192 up_write(&io->io_rwsem);
193 }
194
195 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
196 int rw)
197 {
198 __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
199 }
200
201 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
202 struct inode *inode, struct page *page,
203 nid_t ino, enum page_type type, int rw)
204 {
205 if (has_merged_page(sbi, inode, page, ino, type))
206 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
207 }
208
209 /*
210 * Fill the locked page with data located in the block address.
211 * Return unlocked page.
212 */
213 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
214 {
215 struct bio *bio;
216 struct page *page = fio->encrypted_page ? fio->encrypted_page : fio->page;
217
218 trace_f2fs_submit_page_bio(page, fio);
219 f2fs_trace_ios(fio, 0);
220
221 /* Allocate a new bio */
222 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->rw));
223
224 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
225 bio_put(bio);
226 return -EFAULT;
227 }
228
229 submit_bio(fio->rw, bio);
230 return 0;
231 }
232
233 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
234 {
235 struct f2fs_sb_info *sbi = fio->sbi;
236 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
237 struct f2fs_bio_info *io;
238 bool is_read = is_read_io(fio->rw);
239 struct page *bio_page;
240
241 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
242
243 if (fio->old_blkaddr != NEW_ADDR)
244 verify_block_addr(sbi, fio->old_blkaddr);
245 verify_block_addr(sbi, fio->new_blkaddr);
246
247 down_write(&io->io_rwsem);
248
249 if (!is_read)
250 inc_page_count(sbi, F2FS_WRITEBACK);
251
252 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
253 io->fio.rw != fio->rw))
254 __submit_merged_bio(io);
255 alloc_new:
256 if (io->bio == NULL) {
257 int bio_blocks = MAX_BIO_BLOCKS(sbi);
258
259 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
260 bio_blocks, is_read);
261 io->fio = *fio;
262 }
263
264 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
265
266 if (bio_add_page(io->bio, bio_page, PAGE_CACHE_SIZE, 0) <
267 PAGE_CACHE_SIZE) {
268 __submit_merged_bio(io);
269 goto alloc_new;
270 }
271
272 io->last_block_in_bio = fio->new_blkaddr;
273 f2fs_trace_ios(fio, 0);
274
275 up_write(&io->io_rwsem);
276 trace_f2fs_submit_page_mbio(fio->page, fio);
277 }
278
279 /*
280 * Lock ordering for the change of data block address:
281 * ->data_page
282 * ->node_page
283 * update block addresses in the node page
284 */
285 void set_data_blkaddr(struct dnode_of_data *dn)
286 {
287 struct f2fs_node *rn;
288 __le32 *addr_array;
289 struct page *node_page = dn->node_page;
290 unsigned int ofs_in_node = dn->ofs_in_node;
291
292 f2fs_wait_on_page_writeback(node_page, NODE, true);
293
294 rn = F2FS_NODE(node_page);
295
296 /* Get physical address of data block */
297 addr_array = blkaddr_in_node(rn);
298 addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
299 if (set_page_dirty(node_page))
300 dn->node_changed = true;
301 }
302
303 int reserve_new_block(struct dnode_of_data *dn)
304 {
305 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
306
307 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
308 return -EPERM;
309 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
310 return -ENOSPC;
311
312 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
313
314 dn->data_blkaddr = NEW_ADDR;
315 set_data_blkaddr(dn);
316 mark_inode_dirty(dn->inode);
317 sync_inode_page(dn);
318 return 0;
319 }
320
321 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
322 {
323 bool need_put = dn->inode_page ? false : true;
324 int err;
325
326 err = get_dnode_of_data(dn, index, ALLOC_NODE);
327 if (err)
328 return err;
329
330 if (dn->data_blkaddr == NULL_ADDR)
331 err = reserve_new_block(dn);
332 if (err || need_put)
333 f2fs_put_dnode(dn);
334 return err;
335 }
336
337 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
338 {
339 struct extent_info ei;
340 struct inode *inode = dn->inode;
341
342 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
343 dn->data_blkaddr = ei.blk + index - ei.fofs;
344 return 0;
345 }
346
347 return f2fs_reserve_block(dn, index);
348 }
349
350 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
351 int rw, bool for_write)
352 {
353 struct address_space *mapping = inode->i_mapping;
354 struct dnode_of_data dn;
355 struct page *page;
356 struct extent_info ei;
357 int err;
358 struct f2fs_io_info fio = {
359 .sbi = F2FS_I_SB(inode),
360 .type = DATA,
361 .rw = rw,
362 .encrypted_page = NULL,
363 };
364
365 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
366 return read_mapping_page(mapping, index, NULL);
367
368 page = f2fs_grab_cache_page(mapping, index, for_write);
369 if (!page)
370 return ERR_PTR(-ENOMEM);
371
372 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
373 dn.data_blkaddr = ei.blk + index - ei.fofs;
374 goto got_it;
375 }
376
377 set_new_dnode(&dn, inode, NULL, NULL, 0);
378 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
379 if (err)
380 goto put_err;
381 f2fs_put_dnode(&dn);
382
383 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
384 err = -ENOENT;
385 goto put_err;
386 }
387 got_it:
388 if (PageUptodate(page)) {
389 unlock_page(page);
390 return page;
391 }
392
393 /*
394 * A new dentry page is allocated but not able to be written, since its
395 * new inode page couldn't be allocated due to -ENOSPC.
396 * In such the case, its blkaddr can be remained as NEW_ADDR.
397 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
398 */
399 if (dn.data_blkaddr == NEW_ADDR) {
400 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
401 SetPageUptodate(page);
402 unlock_page(page);
403 return page;
404 }
405
406 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
407 fio.page = page;
408 err = f2fs_submit_page_bio(&fio);
409 if (err)
410 goto put_err;
411 return page;
412
413 put_err:
414 f2fs_put_page(page, 1);
415 return ERR_PTR(err);
416 }
417
418 struct page *find_data_page(struct inode *inode, pgoff_t index)
419 {
420 struct address_space *mapping = inode->i_mapping;
421 struct page *page;
422
423 page = find_get_page(mapping, index);
424 if (page && PageUptodate(page))
425 return page;
426 f2fs_put_page(page, 0);
427
428 page = get_read_data_page(inode, index, READ_SYNC, false);
429 if (IS_ERR(page))
430 return page;
431
432 if (PageUptodate(page))
433 return page;
434
435 wait_on_page_locked(page);
436 if (unlikely(!PageUptodate(page))) {
437 f2fs_put_page(page, 0);
438 return ERR_PTR(-EIO);
439 }
440 return page;
441 }
442
443 /*
444 * If it tries to access a hole, return an error.
445 * Because, the callers, functions in dir.c and GC, should be able to know
446 * whether this page exists or not.
447 */
448 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
449 bool for_write)
450 {
451 struct address_space *mapping = inode->i_mapping;
452 struct page *page;
453 repeat:
454 page = get_read_data_page(inode, index, READ_SYNC, for_write);
455 if (IS_ERR(page))
456 return page;
457
458 /* wait for read completion */
459 lock_page(page);
460 if (unlikely(!PageUptodate(page))) {
461 f2fs_put_page(page, 1);
462 return ERR_PTR(-EIO);
463 }
464 if (unlikely(page->mapping != mapping)) {
465 f2fs_put_page(page, 1);
466 goto repeat;
467 }
468 return page;
469 }
470
471 /*
472 * Caller ensures that this data page is never allocated.
473 * A new zero-filled data page is allocated in the page cache.
474 *
475 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
476 * f2fs_unlock_op().
477 * Note that, ipage is set only by make_empty_dir, and if any error occur,
478 * ipage should be released by this function.
479 */
480 struct page *get_new_data_page(struct inode *inode,
481 struct page *ipage, pgoff_t index, bool new_i_size)
482 {
483 struct address_space *mapping = inode->i_mapping;
484 struct page *page;
485 struct dnode_of_data dn;
486 int err;
487
488 page = f2fs_grab_cache_page(mapping, index, true);
489 if (!page) {
490 /*
491 * before exiting, we should make sure ipage will be released
492 * if any error occur.
493 */
494 f2fs_put_page(ipage, 1);
495 return ERR_PTR(-ENOMEM);
496 }
497
498 set_new_dnode(&dn, inode, ipage, NULL, 0);
499 err = f2fs_reserve_block(&dn, index);
500 if (err) {
501 f2fs_put_page(page, 1);
502 return ERR_PTR(err);
503 }
504 if (!ipage)
505 f2fs_put_dnode(&dn);
506
507 if (PageUptodate(page))
508 goto got_it;
509
510 if (dn.data_blkaddr == NEW_ADDR) {
511 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
512 SetPageUptodate(page);
513 } else {
514 f2fs_put_page(page, 1);
515
516 /* if ipage exists, blkaddr should be NEW_ADDR */
517 f2fs_bug_on(F2FS_I_SB(inode), ipage);
518 page = get_lock_data_page(inode, index, true);
519 if (IS_ERR(page))
520 return page;
521 }
522 got_it:
523 if (new_i_size && i_size_read(inode) <
524 ((loff_t)(index + 1) << PAGE_CACHE_SHIFT)) {
525 i_size_write(inode, ((loff_t)(index + 1) << PAGE_CACHE_SHIFT));
526 /* Only the directory inode sets new_i_size */
527 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
528 }
529 return page;
530 }
531
532 static int __allocate_data_block(struct dnode_of_data *dn)
533 {
534 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
535 struct f2fs_summary sum;
536 struct node_info ni;
537 int seg = CURSEG_WARM_DATA;
538 pgoff_t fofs;
539
540 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
541 return -EPERM;
542
543 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
544 if (dn->data_blkaddr == NEW_ADDR)
545 goto alloc;
546
547 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
548 return -ENOSPC;
549
550 alloc:
551 get_node_info(sbi, dn->nid, &ni);
552 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
553
554 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
555 seg = CURSEG_DIRECT_IO;
556
557 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
558 &sum, seg);
559 set_data_blkaddr(dn);
560
561 /* update i_size */
562 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
563 dn->ofs_in_node;
564 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT))
565 i_size_write(dn->inode,
566 ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT));
567 return 0;
568 }
569
570 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
571 {
572 struct inode *inode = file_inode(iocb->ki_filp);
573 struct f2fs_map_blocks map;
574 ssize_t ret = 0;
575
576 map.m_lblk = F2FS_BYTES_TO_BLK(iocb->ki_pos);
577 map.m_len = F2FS_BLK_ALIGN(iov_iter_count(from));
578 map.m_next_pgofs = NULL;
579
580 if (f2fs_encrypted_inode(inode))
581 return 0;
582
583 if (iocb->ki_flags & IOCB_DIRECT) {
584 ret = f2fs_convert_inline_inode(inode);
585 if (ret)
586 return ret;
587 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
588 }
589 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
590 ret = f2fs_convert_inline_inode(inode);
591 if (ret)
592 return ret;
593 }
594 if (!f2fs_has_inline_data(inode))
595 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
596 return ret;
597 }
598
599 /*
600 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
601 * f2fs_map_blocks structure.
602 * If original data blocks are allocated, then give them to blockdev.
603 * Otherwise,
604 * a. preallocate requested block addresses
605 * b. do not use extent cache for better performance
606 * c. give the block addresses to blockdev
607 */
608 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
609 int create, int flag)
610 {
611 unsigned int maxblocks = map->m_len;
612 struct dnode_of_data dn;
613 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
614 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
615 pgoff_t pgofs, end_offset;
616 int err = 0, ofs = 1;
617 struct extent_info ei;
618 bool allocated = false;
619 block_t blkaddr;
620
621 map->m_len = 0;
622 map->m_flags = 0;
623
624 /* it only supports block size == page size */
625 pgofs = (pgoff_t)map->m_lblk;
626
627 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
628 map->m_pblk = ei.blk + pgofs - ei.fofs;
629 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
630 map->m_flags = F2FS_MAP_MAPPED;
631 goto out;
632 }
633
634 next_dnode:
635 if (create)
636 f2fs_lock_op(sbi);
637
638 /* When reading holes, we need its node page */
639 set_new_dnode(&dn, inode, NULL, NULL, 0);
640 err = get_dnode_of_data(&dn, pgofs, mode);
641 if (err) {
642 if (err == -ENOENT) {
643 err = 0;
644 if (map->m_next_pgofs)
645 *map->m_next_pgofs =
646 get_next_page_offset(&dn, pgofs);
647 }
648 goto unlock_out;
649 }
650
651 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
652
653 next_block:
654 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
655
656 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
657 if (create) {
658 if (unlikely(f2fs_cp_error(sbi))) {
659 err = -EIO;
660 goto sync_out;
661 }
662 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
663 if (blkaddr == NULL_ADDR)
664 err = reserve_new_block(&dn);
665 } else {
666 err = __allocate_data_block(&dn);
667 }
668 if (err)
669 goto sync_out;
670 allocated = true;
671 map->m_flags = F2FS_MAP_NEW;
672 blkaddr = dn.data_blkaddr;
673 } else {
674 if (flag == F2FS_GET_BLOCK_FIEMAP &&
675 blkaddr == NULL_ADDR) {
676 if (map->m_next_pgofs)
677 *map->m_next_pgofs = pgofs + 1;
678 }
679 if (flag != F2FS_GET_BLOCK_FIEMAP ||
680 blkaddr != NEW_ADDR) {
681 if (flag == F2FS_GET_BLOCK_BMAP)
682 err = -ENOENT;
683 goto sync_out;
684 }
685 }
686 }
687
688 if (map->m_len == 0) {
689 /* preallocated unwritten block should be mapped for fiemap. */
690 if (blkaddr == NEW_ADDR)
691 map->m_flags |= F2FS_MAP_UNWRITTEN;
692 map->m_flags |= F2FS_MAP_MAPPED;
693
694 map->m_pblk = blkaddr;
695 map->m_len = 1;
696 } else if ((map->m_pblk != NEW_ADDR &&
697 blkaddr == (map->m_pblk + ofs)) ||
698 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
699 flag == F2FS_GET_BLOCK_PRE_DIO ||
700 flag == F2FS_GET_BLOCK_PRE_AIO) {
701 ofs++;
702 map->m_len++;
703 } else {
704 goto sync_out;
705 }
706
707 dn.ofs_in_node++;
708 pgofs++;
709
710 if (map->m_len < maxblocks) {
711 if (dn.ofs_in_node < end_offset)
712 goto next_block;
713
714 if (allocated)
715 sync_inode_page(&dn);
716 f2fs_put_dnode(&dn);
717
718 if (create) {
719 f2fs_unlock_op(sbi);
720 f2fs_balance_fs(sbi, allocated);
721 }
722 allocated = false;
723 goto next_dnode;
724 }
725
726 sync_out:
727 if (allocated)
728 sync_inode_page(&dn);
729 f2fs_put_dnode(&dn);
730 unlock_out:
731 if (create) {
732 f2fs_unlock_op(sbi);
733 f2fs_balance_fs(sbi, allocated);
734 }
735 out:
736 trace_f2fs_map_blocks(inode, map, err);
737 return err;
738 }
739
740 static int __get_data_block(struct inode *inode, sector_t iblock,
741 struct buffer_head *bh, int create, int flag,
742 pgoff_t *next_pgofs)
743 {
744 struct f2fs_map_blocks map;
745 int ret;
746
747 map.m_lblk = iblock;
748 map.m_len = bh->b_size >> inode->i_blkbits;
749 map.m_next_pgofs = next_pgofs;
750
751 ret = f2fs_map_blocks(inode, &map, create, flag);
752 if (!ret) {
753 map_bh(bh, inode->i_sb, map.m_pblk);
754 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
755 bh->b_size = map.m_len << inode->i_blkbits;
756 }
757 return ret;
758 }
759
760 static int get_data_block(struct inode *inode, sector_t iblock,
761 struct buffer_head *bh_result, int create, int flag,
762 pgoff_t *next_pgofs)
763 {
764 return __get_data_block(inode, iblock, bh_result, create,
765 flag, next_pgofs);
766 }
767
768 static int get_data_block_dio(struct inode *inode, sector_t iblock,
769 struct buffer_head *bh_result, int create)
770 {
771 return __get_data_block(inode, iblock, bh_result, create,
772 F2FS_GET_BLOCK_DIO, NULL);
773 }
774
775 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
776 struct buffer_head *bh_result, int create)
777 {
778 /* Block number less than F2FS MAX BLOCKS */
779 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
780 return -EFBIG;
781
782 return __get_data_block(inode, iblock, bh_result, create,
783 F2FS_GET_BLOCK_BMAP, NULL);
784 }
785
786 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
787 {
788 return (offset >> inode->i_blkbits);
789 }
790
791 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
792 {
793 return (blk << inode->i_blkbits);
794 }
795
796 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
797 u64 start, u64 len)
798 {
799 struct buffer_head map_bh;
800 sector_t start_blk, last_blk;
801 pgoff_t next_pgofs;
802 loff_t isize;
803 u64 logical = 0, phys = 0, size = 0;
804 u32 flags = 0;
805 int ret = 0;
806
807 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
808 if (ret)
809 return ret;
810
811 if (f2fs_has_inline_data(inode)) {
812 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
813 if (ret != -EAGAIN)
814 return ret;
815 }
816
817 inode_lock(inode);
818
819 isize = i_size_read(inode);
820 if (start >= isize)
821 goto out;
822
823 if (start + len > isize)
824 len = isize - start;
825
826 if (logical_to_blk(inode, len) == 0)
827 len = blk_to_logical(inode, 1);
828
829 start_blk = logical_to_blk(inode, start);
830 last_blk = logical_to_blk(inode, start + len - 1);
831
832 next:
833 memset(&map_bh, 0, sizeof(struct buffer_head));
834 map_bh.b_size = len;
835
836 ret = get_data_block(inode, start_blk, &map_bh, 0,
837 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
838 if (ret)
839 goto out;
840
841 /* HOLE */
842 if (!buffer_mapped(&map_bh)) {
843 start_blk = next_pgofs;
844 /* Go through holes util pass the EOF */
845 if (blk_to_logical(inode, start_blk) < isize)
846 goto prep_next;
847 /* Found a hole beyond isize means no more extents.
848 * Note that the premise is that filesystems don't
849 * punch holes beyond isize and keep size unchanged.
850 */
851 flags |= FIEMAP_EXTENT_LAST;
852 }
853
854 if (size) {
855 if (f2fs_encrypted_inode(inode))
856 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
857
858 ret = fiemap_fill_next_extent(fieinfo, logical,
859 phys, size, flags);
860 }
861
862 if (start_blk > last_blk || ret)
863 goto out;
864
865 logical = blk_to_logical(inode, start_blk);
866 phys = blk_to_logical(inode, map_bh.b_blocknr);
867 size = map_bh.b_size;
868 flags = 0;
869 if (buffer_unwritten(&map_bh))
870 flags = FIEMAP_EXTENT_UNWRITTEN;
871
872 start_blk += logical_to_blk(inode, size);
873
874 prep_next:
875 cond_resched();
876 if (fatal_signal_pending(current))
877 ret = -EINTR;
878 else
879 goto next;
880 out:
881 if (ret == 1)
882 ret = 0;
883
884 inode_unlock(inode);
885 return ret;
886 }
887
888 /*
889 * This function was originally taken from fs/mpage.c, and customized for f2fs.
890 * Major change was from block_size == page_size in f2fs by default.
891 */
892 static int f2fs_mpage_readpages(struct address_space *mapping,
893 struct list_head *pages, struct page *page,
894 unsigned nr_pages)
895 {
896 struct bio *bio = NULL;
897 unsigned page_idx;
898 sector_t last_block_in_bio = 0;
899 struct inode *inode = mapping->host;
900 const unsigned blkbits = inode->i_blkbits;
901 const unsigned blocksize = 1 << blkbits;
902 sector_t block_in_file;
903 sector_t last_block;
904 sector_t last_block_in_file;
905 sector_t block_nr;
906 struct block_device *bdev = inode->i_sb->s_bdev;
907 struct f2fs_map_blocks map;
908
909 map.m_pblk = 0;
910 map.m_lblk = 0;
911 map.m_len = 0;
912 map.m_flags = 0;
913 map.m_next_pgofs = NULL;
914
915 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
916
917 prefetchw(&page->flags);
918 if (pages) {
919 page = list_entry(pages->prev, struct page, lru);
920 list_del(&page->lru);
921 if (add_to_page_cache_lru(page, mapping,
922 page->index, GFP_KERNEL))
923 goto next_page;
924 }
925
926 block_in_file = (sector_t)page->index;
927 last_block = block_in_file + nr_pages;
928 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
929 blkbits;
930 if (last_block > last_block_in_file)
931 last_block = last_block_in_file;
932
933 /*
934 * Map blocks using the previous result first.
935 */
936 if ((map.m_flags & F2FS_MAP_MAPPED) &&
937 block_in_file > map.m_lblk &&
938 block_in_file < (map.m_lblk + map.m_len))
939 goto got_it;
940
941 /*
942 * Then do more f2fs_map_blocks() calls until we are
943 * done with this page.
944 */
945 map.m_flags = 0;
946
947 if (block_in_file < last_block) {
948 map.m_lblk = block_in_file;
949 map.m_len = last_block - block_in_file;
950
951 if (f2fs_map_blocks(inode, &map, 0,
952 F2FS_GET_BLOCK_READ))
953 goto set_error_page;
954 }
955 got_it:
956 if ((map.m_flags & F2FS_MAP_MAPPED)) {
957 block_nr = map.m_pblk + block_in_file - map.m_lblk;
958 SetPageMappedToDisk(page);
959
960 if (!PageUptodate(page) && !cleancache_get_page(page)) {
961 SetPageUptodate(page);
962 goto confused;
963 }
964 } else {
965 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
966 SetPageUptodate(page);
967 unlock_page(page);
968 goto next_page;
969 }
970
971 /*
972 * This page will go to BIO. Do we need to send this
973 * BIO off first?
974 */
975 if (bio && (last_block_in_bio != block_nr - 1)) {
976 submit_and_realloc:
977 submit_bio(READ, bio);
978 bio = NULL;
979 }
980 if (bio == NULL) {
981 struct f2fs_crypto_ctx *ctx = NULL;
982
983 if (f2fs_encrypted_inode(inode) &&
984 S_ISREG(inode->i_mode)) {
985
986 ctx = f2fs_get_crypto_ctx(inode);
987 if (IS_ERR(ctx))
988 goto set_error_page;
989
990 /* wait the page to be moved by cleaning */
991 f2fs_wait_on_encrypted_page_writeback(
992 F2FS_I_SB(inode), block_nr);
993 }
994
995 bio = bio_alloc(GFP_KERNEL,
996 min_t(int, nr_pages, BIO_MAX_PAGES));
997 if (!bio) {
998 if (ctx)
999 f2fs_release_crypto_ctx(ctx);
1000 goto set_error_page;
1001 }
1002 bio->bi_bdev = bdev;
1003 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1004 bio->bi_end_io = f2fs_read_end_io;
1005 bio->bi_private = ctx;
1006 }
1007
1008 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1009 goto submit_and_realloc;
1010
1011 last_block_in_bio = block_nr;
1012 goto next_page;
1013 set_error_page:
1014 SetPageError(page);
1015 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1016 unlock_page(page);
1017 goto next_page;
1018 confused:
1019 if (bio) {
1020 submit_bio(READ, bio);
1021 bio = NULL;
1022 }
1023 unlock_page(page);
1024 next_page:
1025 if (pages)
1026 page_cache_release(page);
1027 }
1028 BUG_ON(pages && !list_empty(pages));
1029 if (bio)
1030 submit_bio(READ, bio);
1031 return 0;
1032 }
1033
1034 static int f2fs_read_data_page(struct file *file, struct page *page)
1035 {
1036 struct inode *inode = page->mapping->host;
1037 int ret = -EAGAIN;
1038
1039 trace_f2fs_readpage(page, DATA);
1040
1041 /* If the file has inline data, try to read it directly */
1042 if (f2fs_has_inline_data(inode))
1043 ret = f2fs_read_inline_data(inode, page);
1044 if (ret == -EAGAIN)
1045 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1046 return ret;
1047 }
1048
1049 static int f2fs_read_data_pages(struct file *file,
1050 struct address_space *mapping,
1051 struct list_head *pages, unsigned nr_pages)
1052 {
1053 struct inode *inode = file->f_mapping->host;
1054 struct page *page = list_entry(pages->prev, struct page, lru);
1055
1056 trace_f2fs_readpages(inode, page, nr_pages);
1057
1058 /* If the file has inline data, skip readpages */
1059 if (f2fs_has_inline_data(inode))
1060 return 0;
1061
1062 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1063 }
1064
1065 int do_write_data_page(struct f2fs_io_info *fio)
1066 {
1067 struct page *page = fio->page;
1068 struct inode *inode = page->mapping->host;
1069 struct dnode_of_data dn;
1070 int err = 0;
1071
1072 set_new_dnode(&dn, inode, NULL, NULL, 0);
1073 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1074 if (err)
1075 return err;
1076
1077 fio->old_blkaddr = dn.data_blkaddr;
1078
1079 /* This page is already truncated */
1080 if (fio->old_blkaddr == NULL_ADDR) {
1081 ClearPageUptodate(page);
1082 goto out_writepage;
1083 }
1084
1085 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1086
1087 /* wait for GCed encrypted page writeback */
1088 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1089 fio->old_blkaddr);
1090
1091 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1092 if (IS_ERR(fio->encrypted_page)) {
1093 err = PTR_ERR(fio->encrypted_page);
1094 goto out_writepage;
1095 }
1096 }
1097
1098 set_page_writeback(page);
1099
1100 /*
1101 * If current allocation needs SSR,
1102 * it had better in-place writes for updated data.
1103 */
1104 if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1105 !is_cold_data(page) &&
1106 !IS_ATOMIC_WRITTEN_PAGE(page) &&
1107 need_inplace_update(inode))) {
1108 rewrite_data_page(fio);
1109 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1110 trace_f2fs_do_write_data_page(page, IPU);
1111 } else {
1112 write_data_page(&dn, fio);
1113 set_data_blkaddr(&dn);
1114 f2fs_update_extent_cache(&dn);
1115 trace_f2fs_do_write_data_page(page, OPU);
1116 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1117 if (page->index == 0)
1118 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1119 }
1120 out_writepage:
1121 f2fs_put_dnode(&dn);
1122 return err;
1123 }
1124
1125 static int f2fs_write_data_page(struct page *page,
1126 struct writeback_control *wbc)
1127 {
1128 struct inode *inode = page->mapping->host;
1129 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1130 loff_t i_size = i_size_read(inode);
1131 const pgoff_t end_index = ((unsigned long long) i_size)
1132 >> PAGE_CACHE_SHIFT;
1133 unsigned offset = 0;
1134 bool need_balance_fs = false;
1135 int err = 0;
1136 struct f2fs_io_info fio = {
1137 .sbi = sbi,
1138 .type = DATA,
1139 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1140 .page = page,
1141 .encrypted_page = NULL,
1142 };
1143
1144 trace_f2fs_writepage(page, DATA);
1145
1146 if (page->index < end_index)
1147 goto write;
1148
1149 /*
1150 * If the offset is out-of-range of file size,
1151 * this page does not have to be written to disk.
1152 */
1153 offset = i_size & (PAGE_CACHE_SIZE - 1);
1154 if ((page->index >= end_index + 1) || !offset)
1155 goto out;
1156
1157 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1158 write:
1159 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1160 goto redirty_out;
1161 if (f2fs_is_drop_cache(inode))
1162 goto out;
1163 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1164 available_free_memory(sbi, BASE_CHECK))
1165 goto redirty_out;
1166
1167 /* Dentry blocks are controlled by checkpoint */
1168 if (S_ISDIR(inode->i_mode)) {
1169 if (unlikely(f2fs_cp_error(sbi)))
1170 goto redirty_out;
1171 err = do_write_data_page(&fio);
1172 goto done;
1173 }
1174
1175 /* we should bypass data pages to proceed the kworkder jobs */
1176 if (unlikely(f2fs_cp_error(sbi))) {
1177 SetPageError(page);
1178 goto out;
1179 }
1180
1181 if (!wbc->for_reclaim)
1182 need_balance_fs = true;
1183 else if (has_not_enough_free_secs(sbi, 0))
1184 goto redirty_out;
1185
1186 err = -EAGAIN;
1187 f2fs_lock_op(sbi);
1188 if (f2fs_has_inline_data(inode))
1189 err = f2fs_write_inline_data(inode, page);
1190 if (err == -EAGAIN)
1191 err = do_write_data_page(&fio);
1192 f2fs_unlock_op(sbi);
1193 done:
1194 if (err && err != -ENOENT)
1195 goto redirty_out;
1196
1197 clear_cold_data(page);
1198 out:
1199 inode_dec_dirty_pages(inode);
1200 if (err)
1201 ClearPageUptodate(page);
1202
1203 if (wbc->for_reclaim) {
1204 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1205 remove_dirty_inode(inode);
1206 }
1207
1208 unlock_page(page);
1209 f2fs_balance_fs(sbi, need_balance_fs);
1210
1211 if (unlikely(f2fs_cp_error(sbi)))
1212 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1213
1214 return 0;
1215
1216 redirty_out:
1217 redirty_page_for_writepage(wbc, page);
1218 return AOP_WRITEPAGE_ACTIVATE;
1219 }
1220
1221 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1222 void *data)
1223 {
1224 struct address_space *mapping = data;
1225 int ret = mapping->a_ops->writepage(page, wbc);
1226 mapping_set_error(mapping, ret);
1227 return ret;
1228 }
1229
1230 /*
1231 * This function was copied from write_cche_pages from mm/page-writeback.c.
1232 * The major change is making write step of cold data page separately from
1233 * warm/hot data page.
1234 */
1235 static int f2fs_write_cache_pages(struct address_space *mapping,
1236 struct writeback_control *wbc, writepage_t writepage,
1237 void *data)
1238 {
1239 int ret = 0;
1240 int done = 0;
1241 struct pagevec pvec;
1242 int nr_pages;
1243 pgoff_t uninitialized_var(writeback_index);
1244 pgoff_t index;
1245 pgoff_t end; /* Inclusive */
1246 pgoff_t done_index;
1247 int cycled;
1248 int range_whole = 0;
1249 int tag;
1250 int step = 0;
1251
1252 pagevec_init(&pvec, 0);
1253 next:
1254 if (wbc->range_cyclic) {
1255 writeback_index = mapping->writeback_index; /* prev offset */
1256 index = writeback_index;
1257 if (index == 0)
1258 cycled = 1;
1259 else
1260 cycled = 0;
1261 end = -1;
1262 } else {
1263 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1264 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1265 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1266 range_whole = 1;
1267 cycled = 1; /* ignore range_cyclic tests */
1268 }
1269 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1270 tag = PAGECACHE_TAG_TOWRITE;
1271 else
1272 tag = PAGECACHE_TAG_DIRTY;
1273 retry:
1274 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1275 tag_pages_for_writeback(mapping, index, end);
1276 done_index = index;
1277 while (!done && (index <= end)) {
1278 int i;
1279
1280 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1281 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1282 if (nr_pages == 0)
1283 break;
1284
1285 for (i = 0; i < nr_pages; i++) {
1286 struct page *page = pvec.pages[i];
1287
1288 if (page->index > end) {
1289 done = 1;
1290 break;
1291 }
1292
1293 done_index = page->index;
1294
1295 lock_page(page);
1296
1297 if (unlikely(page->mapping != mapping)) {
1298 continue_unlock:
1299 unlock_page(page);
1300 continue;
1301 }
1302
1303 if (!PageDirty(page)) {
1304 /* someone wrote it for us */
1305 goto continue_unlock;
1306 }
1307
1308 if (step == is_cold_data(page))
1309 goto continue_unlock;
1310
1311 if (PageWriteback(page)) {
1312 if (wbc->sync_mode != WB_SYNC_NONE)
1313 f2fs_wait_on_page_writeback(page,
1314 DATA, true);
1315 else
1316 goto continue_unlock;
1317 }
1318
1319 BUG_ON(PageWriteback(page));
1320 if (!clear_page_dirty_for_io(page))
1321 goto continue_unlock;
1322
1323 ret = (*writepage)(page, wbc, data);
1324 if (unlikely(ret)) {
1325 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1326 unlock_page(page);
1327 ret = 0;
1328 } else {
1329 done_index = page->index + 1;
1330 done = 1;
1331 break;
1332 }
1333 }
1334
1335 if (--wbc->nr_to_write <= 0 &&
1336 wbc->sync_mode == WB_SYNC_NONE) {
1337 done = 1;
1338 break;
1339 }
1340 }
1341 pagevec_release(&pvec);
1342 cond_resched();
1343 }
1344
1345 if (step < 1) {
1346 step++;
1347 goto next;
1348 }
1349
1350 if (!cycled && !done) {
1351 cycled = 1;
1352 index = 0;
1353 end = writeback_index - 1;
1354 goto retry;
1355 }
1356 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1357 mapping->writeback_index = done_index;
1358
1359 return ret;
1360 }
1361
1362 static int f2fs_write_data_pages(struct address_space *mapping,
1363 struct writeback_control *wbc)
1364 {
1365 struct inode *inode = mapping->host;
1366 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1367 bool locked = false;
1368 int ret;
1369 long diff;
1370
1371 /* deal with chardevs and other special file */
1372 if (!mapping->a_ops->writepage)
1373 return 0;
1374
1375 /* skip writing if there is no dirty page in this inode */
1376 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1377 return 0;
1378
1379 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1380 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1381 available_free_memory(sbi, DIRTY_DENTS))
1382 goto skip_write;
1383
1384 /* skip writing during file defragment */
1385 if (is_inode_flag_set(F2FS_I(inode), FI_DO_DEFRAG))
1386 goto skip_write;
1387
1388 /* during POR, we don't need to trigger writepage at all. */
1389 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1390 goto skip_write;
1391
1392 trace_f2fs_writepages(mapping->host, wbc, DATA);
1393
1394 diff = nr_pages_to_write(sbi, DATA, wbc);
1395
1396 if (!S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_ALL) {
1397 mutex_lock(&sbi->writepages);
1398 locked = true;
1399 }
1400 ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1401 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
1402 if (locked)
1403 mutex_unlock(&sbi->writepages);
1404
1405 remove_dirty_inode(inode);
1406
1407 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1408 return ret;
1409
1410 skip_write:
1411 wbc->pages_skipped += get_dirty_pages(inode);
1412 trace_f2fs_writepages(mapping->host, wbc, DATA);
1413 return 0;
1414 }
1415
1416 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1417 {
1418 struct inode *inode = mapping->host;
1419 loff_t i_size = i_size_read(inode);
1420
1421 if (to > i_size) {
1422 truncate_pagecache(inode, i_size);
1423 truncate_blocks(inode, i_size, true);
1424 }
1425 }
1426
1427 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1428 struct page *page, loff_t pos, unsigned len,
1429 block_t *blk_addr, bool *node_changed)
1430 {
1431 struct inode *inode = page->mapping->host;
1432 pgoff_t index = page->index;
1433 struct dnode_of_data dn;
1434 struct page *ipage;
1435 bool locked = false;
1436 struct extent_info ei;
1437 int err = 0;
1438
1439 /*
1440 * we already allocated all the blocks, so we don't need to get
1441 * the block addresses when there is no need to fill the page.
1442 */
1443 if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1444 len == PAGE_CACHE_SIZE)
1445 return 0;
1446
1447 if (f2fs_has_inline_data(inode) ||
1448 (pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1449 f2fs_lock_op(sbi);
1450 locked = true;
1451 }
1452 restart:
1453 /* check inline_data */
1454 ipage = get_node_page(sbi, inode->i_ino);
1455 if (IS_ERR(ipage)) {
1456 err = PTR_ERR(ipage);
1457 goto unlock_out;
1458 }
1459
1460 set_new_dnode(&dn, inode, ipage, ipage, 0);
1461
1462 if (f2fs_has_inline_data(inode)) {
1463 if (pos + len <= MAX_INLINE_DATA) {
1464 read_inline_data(page, ipage);
1465 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1466 set_inline_node(ipage);
1467 } else {
1468 err = f2fs_convert_inline_page(&dn, page);
1469 if (err)
1470 goto out;
1471 if (dn.data_blkaddr == NULL_ADDR)
1472 err = f2fs_get_block(&dn, index);
1473 }
1474 } else if (locked) {
1475 err = f2fs_get_block(&dn, index);
1476 } else {
1477 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1478 dn.data_blkaddr = ei.blk + index - ei.fofs;
1479 } else {
1480 /* hole case */
1481 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1482 if (err || (!err && dn.data_blkaddr == NULL_ADDR)) {
1483 f2fs_put_dnode(&dn);
1484 f2fs_lock_op(sbi);
1485 locked = true;
1486 goto restart;
1487 }
1488 }
1489 }
1490
1491 /* convert_inline_page can make node_changed */
1492 *blk_addr = dn.data_blkaddr;
1493 *node_changed = dn.node_changed;
1494 out:
1495 f2fs_put_dnode(&dn);
1496 unlock_out:
1497 if (locked)
1498 f2fs_unlock_op(sbi);
1499 return err;
1500 }
1501
1502 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1503 loff_t pos, unsigned len, unsigned flags,
1504 struct page **pagep, void **fsdata)
1505 {
1506 struct inode *inode = mapping->host;
1507 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1508 struct page *page = NULL;
1509 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1510 bool need_balance = false;
1511 block_t blkaddr = NULL_ADDR;
1512 int err = 0;
1513
1514 trace_f2fs_write_begin(inode, pos, len, flags);
1515
1516 /*
1517 * We should check this at this moment to avoid deadlock on inode page
1518 * and #0 page. The locking rule for inline_data conversion should be:
1519 * lock_page(page #0) -> lock_page(inode_page)
1520 */
1521 if (index != 0) {
1522 err = f2fs_convert_inline_inode(inode);
1523 if (err)
1524 goto fail;
1525 }
1526 repeat:
1527 page = grab_cache_page_write_begin(mapping, index, flags);
1528 if (!page) {
1529 err = -ENOMEM;
1530 goto fail;
1531 }
1532
1533 *pagep = page;
1534
1535 err = prepare_write_begin(sbi, page, pos, len,
1536 &blkaddr, &need_balance);
1537 if (err)
1538 goto fail;
1539
1540 if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1541 unlock_page(page);
1542 f2fs_balance_fs(sbi, true);
1543 lock_page(page);
1544 if (page->mapping != mapping) {
1545 /* The page got truncated from under us */
1546 f2fs_put_page(page, 1);
1547 goto repeat;
1548 }
1549 }
1550
1551 f2fs_wait_on_page_writeback(page, DATA, false);
1552
1553 /* wait for GCed encrypted page writeback */
1554 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1555 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1556
1557 if (len == PAGE_CACHE_SIZE)
1558 goto out_update;
1559 if (PageUptodate(page))
1560 goto out_clear;
1561
1562 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1563 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1564 unsigned end = start + len;
1565
1566 /* Reading beyond i_size is simple: memset to zero */
1567 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1568 goto out_update;
1569 }
1570
1571 if (blkaddr == NEW_ADDR) {
1572 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1573 } else {
1574 struct f2fs_io_info fio = {
1575 .sbi = sbi,
1576 .type = DATA,
1577 .rw = READ_SYNC,
1578 .old_blkaddr = blkaddr,
1579 .new_blkaddr = blkaddr,
1580 .page = page,
1581 .encrypted_page = NULL,
1582 };
1583 err = f2fs_submit_page_bio(&fio);
1584 if (err)
1585 goto fail;
1586
1587 lock_page(page);
1588 if (unlikely(!PageUptodate(page))) {
1589 err = -EIO;
1590 goto fail;
1591 }
1592 if (unlikely(page->mapping != mapping)) {
1593 f2fs_put_page(page, 1);
1594 goto repeat;
1595 }
1596
1597 /* avoid symlink page */
1598 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1599 err = f2fs_decrypt(page);
1600 if (err)
1601 goto fail;
1602 }
1603 }
1604 out_update:
1605 SetPageUptodate(page);
1606 out_clear:
1607 clear_cold_data(page);
1608 return 0;
1609
1610 fail:
1611 f2fs_put_page(page, 1);
1612 f2fs_write_failed(mapping, pos + len);
1613 return err;
1614 }
1615
1616 static int f2fs_write_end(struct file *file,
1617 struct address_space *mapping,
1618 loff_t pos, unsigned len, unsigned copied,
1619 struct page *page, void *fsdata)
1620 {
1621 struct inode *inode = page->mapping->host;
1622
1623 trace_f2fs_write_end(inode, pos, len, copied);
1624
1625 set_page_dirty(page);
1626
1627 if (pos + copied > i_size_read(inode)) {
1628 i_size_write(inode, pos + copied);
1629 mark_inode_dirty(inode);
1630 }
1631
1632 f2fs_put_page(page, 1);
1633 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1634 return copied;
1635 }
1636
1637 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1638 loff_t offset)
1639 {
1640 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1641
1642 if (offset & blocksize_mask)
1643 return -EINVAL;
1644
1645 if (iov_iter_alignment(iter) & blocksize_mask)
1646 return -EINVAL;
1647
1648 return 0;
1649 }
1650
1651 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1652 loff_t offset)
1653 {
1654 struct address_space *mapping = iocb->ki_filp->f_mapping;
1655 struct inode *inode = mapping->host;
1656 size_t count = iov_iter_count(iter);
1657 int err;
1658
1659 err = check_direct_IO(inode, iter, offset);
1660 if (err)
1661 return err;
1662
1663 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1664 return 0;
1665
1666 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1667
1668 err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1669 if (err < 0 && iov_iter_rw(iter) == WRITE)
1670 f2fs_write_failed(mapping, offset + count);
1671
1672 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1673
1674 return err;
1675 }
1676
1677 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1678 unsigned int length)
1679 {
1680 struct inode *inode = page->mapping->host;
1681 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1682
1683 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1684 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1685 return;
1686
1687 if (PageDirty(page)) {
1688 if (inode->i_ino == F2FS_META_INO(sbi))
1689 dec_page_count(sbi, F2FS_DIRTY_META);
1690 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1691 dec_page_count(sbi, F2FS_DIRTY_NODES);
1692 else
1693 inode_dec_dirty_pages(inode);
1694 }
1695
1696 /* This is atomic written page, keep Private */
1697 if (IS_ATOMIC_WRITTEN_PAGE(page))
1698 return;
1699
1700 ClearPagePrivate(page);
1701 }
1702
1703 int f2fs_release_page(struct page *page, gfp_t wait)
1704 {
1705 /* If this is dirty page, keep PagePrivate */
1706 if (PageDirty(page))
1707 return 0;
1708
1709 /* This is atomic written page, keep Private */
1710 if (IS_ATOMIC_WRITTEN_PAGE(page))
1711 return 0;
1712
1713 ClearPagePrivate(page);
1714 return 1;
1715 }
1716
1717 static int f2fs_set_data_page_dirty(struct page *page)
1718 {
1719 struct address_space *mapping = page->mapping;
1720 struct inode *inode = mapping->host;
1721
1722 trace_f2fs_set_page_dirty(page, DATA);
1723
1724 SetPageUptodate(page);
1725
1726 if (f2fs_is_atomic_file(inode)) {
1727 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1728 register_inmem_page(inode, page);
1729 return 1;
1730 }
1731 /*
1732 * Previously, this page has been registered, we just
1733 * return here.
1734 */
1735 return 0;
1736 }
1737
1738 if (!PageDirty(page)) {
1739 __set_page_dirty_nobuffers(page);
1740 update_dirty_page(inode, page);
1741 return 1;
1742 }
1743 return 0;
1744 }
1745
1746 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1747 {
1748 struct inode *inode = mapping->host;
1749
1750 if (f2fs_has_inline_data(inode))
1751 return 0;
1752
1753 /* make sure allocating whole blocks */
1754 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1755 filemap_write_and_wait(mapping);
1756
1757 return generic_block_bmap(mapping, block, get_data_block_bmap);
1758 }
1759
1760 const struct address_space_operations f2fs_dblock_aops = {
1761 .readpage = f2fs_read_data_page,
1762 .readpages = f2fs_read_data_pages,
1763 .writepage = f2fs_write_data_page,
1764 .writepages = f2fs_write_data_pages,
1765 .write_begin = f2fs_write_begin,
1766 .write_end = f2fs_write_end,
1767 .set_page_dirty = f2fs_set_data_page_dirty,
1768 .invalidatepage = f2fs_invalidate_page,
1769 .releasepage = f2fs_release_page,
1770 .direct_IO = f2fs_direct_IO,
1771 .bmap = f2fs_bmap,
1772 };
Linux kernel - Deliverables
This page took 0.070786 seconds and 5 git commands to generate.